Joe Landman, on 01/28/2010 06:01 PM wrote:
Pasi Kärkkäinen wrote:
Hello list,
Please check these news items:
http://blog.fosketts.net/2010/01/14/microsoft-intel-push-million-iscsi-iops/
http://communities.intel.com/community/openportit/server/blog/2010/01/19/1000000-iops-with-iscsi--thats-not-a-typo
http://www.infostor.com/index/blogs_new/dave_simpson_storage/blogs/infostor/dave_simpon_storage/post987_37501094375591341.html
"1,030,000 IOPS over a single 10 Gb Ethernet link"
This is less than 1us per IOP. Interesting. Their hardware may not
actually support this. 10GbE typically is 7-10us, though ConnectX and
some others get down to 2ish.
You can't calculate any latency for this test as 1/IOPS number.
Otherwise you can measure that a half a second satellite link has lower
latency, than your local 100Mbps net ;).
You know, each link has 2 properties: latency and bandwidth. They are
*independent* (or orthogonal), this is fundamental. This means that if
you know one parameter, you can't figure out another one.
So, if you send single 1 byte IO at time, 1/IOPS will mean latency (I
suppose 1/bandwidth << 1/IOPS, otherwise latency is 1/IOPS -
1/bandwidth), but if you fully fill the link, i.e. have at least
latency*bandwidth data in-flight, with 1 byte IOs 1/IOPS will mean time
to transfer each IO, i.e. 1/bandwidth.
Hence, tests which fully fill the link, as in this case, are latency
insensitive and only test bandwidth, so you can't figure out any latency
from them. You will get the same IOPS results (millions, possibly) for
any link's latency, even if it's hundreds of seconds.
Initiator and target systems can also be considered as "links" with the
same 2 independent parameters: latency (time to process each IO) and
bandwidth (how many IOs at time can be processed), where bandwidth !=
latency * CPUs_num, because SMP systems don't scale linearly. Plus, in
this test there are many targets working in parallel.
Thus, in this test IOs sent through many links, including the above
"links", and fully filled them. So, for this test 1/IOPS is close to
IO_size/the_worst_link_bandwidth.
Vlad
--
You received this message because you are subscribed to the Google Groups
"open-iscsi" group.
To post to this group, send email to [email protected].
To unsubscribe from this group, send email to
[email protected].
For more options, visit this group at
http://groups.google.com/group/open-iscsi?hl=en.
